Re: [Xen-devel] [PATCH v2 08/15] xen/arm: vgic-v3: Emulate correctly the re-distributor

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Tue, 2015-02-03 at 13:13 +0000, Julien Grall wrote:
> My point was, we have to trust a bit the device tree given by the
> platform. If the device tree is buggy, then bare-metal linux will
> unlikely boot.

I think overlapping or not correctly size aligned, assuming such goes
against the binding == buggy device tree, which we can decide to either
proactively test for it and warn, or just trust not to be the case and
damn the consequences on buggy platforms, that's fine (although
pragmatically, we might want to support a buggy platform at some
point...).

But we initially started off talking about the ordering of the entries.
It's not clear to me what the requirements from the spec (ePAPR, the
bindings et al) are in this regard, so I don't know that we can simply
claim device trees which do not sort them are buggy, and therefore
whether we can trust them to be ordered or not.

If something requires that the rdist regions are ordered then we can
trust that and treat it like overlap or sizing/alignment etc. If the
spec doesn't say anything about the entries being ordered in some
particular way then we should try not to assume it or trust them to be
that way.

(AFAICT our current handling of the memory nodes violates this, at least
I can't find anything in ePAPR which requires an ordering but we do).

Anyway...

> But I agree, that we have to sort the region of re-distributors at init.

That would be one way to avoid trawling through all the specs etc ;-)

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel